An AI backend that archived itself into a 200KB successor

What it does Unbody was pitched as a modular, open-source backend for AI-native software — designed for knowledge and ingestion rather than static relational data. Per its own topics and the collapsed setup notes, it ran a Dockerized stack with Temporal for indexing pipelines, a CLI for adding data sources, and hooks for OpenAI. The README is light on architecture details, but the tags make the intended use cases clear: RAG, chatbots, and vector-backed knowledge bases.

The interesting bit The project did not die; it dieted. The team archived the monolithic repository and channeled the same mission into Adapt, a sub-200KB provider-agnostic memory framework that runs in the browser or on a server. That is a dramatic shrinkage in both scope and binary size.

Key highlights

• Explicitly archived and no longer actively maintained.
• Originally targeted data ingestion, vector search, and generative-AI backends.
• Required Docker, Temporal, yarn, and an OpenAI key — a fairly heavy toolchain.
• The spiritual successor, Adapt, claims to be under 200KB and provider-agnostic.

Caveats

• The README is vague on actual features; most operational details are buried in collapsed setup instructions.
• Since it is archived, there is no active development or support.

Verdict Treat this repo as a historical signpost and follow the link to Adapt instead. If you need a drop-in backend alternative for AI apps, the README makes clear you will not find it here anymore.